Advances in Data Management Systems

Project Proposal

 

Project Team Info:

            Nicolas Bonvin (nicolas.bonvin AT epfl.ch)

            Surender Yerva (surenderreddy.yerva AT epfl.ch)

             Rammohan Narendula (rammohan.narendula AT epfl.ch)

Project Web Page:

http://lsirpeople.epfl.ch/narendul/adb/StagedDB.html

 
Project Description: 

Scheduling in StagedDB: StagedDB is a DBMS software, which is designed and realized as multiple modules. These multiple modules can be grouped into self-contained stages connected to each other through the queues. Each of this stage can contain multiple threads. Each bigger and complex stage (such as execution engine) can be thought of collection of a number of micro engines (say one engine per operator). Each micro engine works with certain part of the query execution- like sort, join, scan from disk. 

There are two levels of scheduling involved in such a setting: one is scheduling at stages level and the other is scheduling with in a stage across micro engines. As a part of this project we would like to design and evaluate various scheduling policies with in a stage- namely execution stage. 

Different Possible Scheduling Policies across Micro Engines

  1. Round-Robin
  2. Random
  3. FCFS
  4. Queue size based
  5. Few more if time permits
Different Possible Scheduling Policies with in Micro Engines
  1. FCFS
  2. Random
  3. Few more if time permits (Ex: based on overlapped factor)
We basically design a simulation framework in MATLAB/JAVA to compare all the policies.

Possible planned performance metrics include
  1. Number of disk I/Os 
  2. Throughput
  3. Response time
  4. (If possible) the maximum amount of memory consumed for the same load
against
  1. Time difference between incoming queries

Logistics:

Plan of Attack and Schedule:

Tentative Schedule:

 Milestones:

Tentative Goal:

        Ready with design for the simulations.

        Decent progress in the simulations implementation.

Literature Search:

Resources Needed:

Getting Started:

So far we have gone through the suggested papers. We identified the scheduling policies which are to be evaluated and collected few metrics to be studied. 

We got decent understanding of the model and about to start.